Everything we know, written down.
- White paper
Inspectable incident reconstruction
The method: time normalization, named rules, four claim states.
Read → - Architecture
How it is built
Pipeline, rules, API, storage, and the decisions behind them.
Open on GitHub → - Security
Controls and limits
What the demo protects, what it does not, what a real deployment needs.
Open on GitHub → - Demo case
CASE-001, fully specified
Six evidence items, expected answers, acceptance checks.
Open on GitHub → - Evaluation
How we verify it works
Known-answer tests, blind case, determinism, clean setup.
Open on GitHub → - API
API reference
Every endpoint, request and error shape.
Open on GitHub →
Questions judges ask
Is this AI?
No model decides anything. Rules are fixed code with IDs; the same input always gives the same output.
Is the data real?
No. Every case in the demo is fictional and labeled as such on every screen.
Does it accuse anyone?
Never. Identity stays Unresolved unless a source supports it.
What is simulated?
Video, audio and PDF extraction comes from metadata in the MVP, and the screen says so. Log files are parsed for real.
How is a clock offset found?
From anchors: moments a camera shares with sources on a trusted clock. The median difference is the offset, and every pair is shown.